mft {paltran}R Documentation

multinomial transfer function for Palaeolimnology

Description

Here the species abundances were assigned to different abundance classes, for each class a species response curve is generated using GLM. These response curves are used to infer the environmental parameter of interest (s. Details)

Usage

mft(..., d.plot = TRUE, scale = TRUE, datatype = TRUE,
        min.occ = 4,set.zero = FALSE, not.av = c("zero", "lower", "max"),
        val = "loo",class=c(1,5,10,30,60),out=TRUE)

Arguments

... x,y,z: required: species training set (x) as matrix and related environmental parameter (y). optional: core data (z) - species data from a sediment core
d.plot if TRUE diagnostic plots are given at the end of the analysis
scale should the data scaled up to 100 percent? (Default)
datatype What kind of data input? TRUE for percentage data, otherwise no transformation to class date will bee done
min.occ minimum number of species occurrence within on class
set.zero If a species occurs within on class less than n time the values will be set as zero, if set.zero = TRUE. If set.zero = FALSE, the values will be allocate to the next lower class
not.av If a species occurs within a abundance class that is not given in the training set, three possibilities are available. "zero" means that this species will be excluded, "lower" stands for to use the next lower class density. If "max" is chosen, the cumulative species response curve of all classes will be calculate and use
val validation method: So fare only Leave-one-out is performed
class what classes should be used? Default is 0,1,5,10,30,60
out should the result bee showen on the console?

Details

The relative abundances of a taxon within a data set is transformed to abundance classes (class 1: 0<x<=1,class 2: 1<x<=5, class 3: 5<x<=10, class 4 10<x<=30, class 5 30<x<=60, class 6: 60-100 percent abundance, number of classes and class borders can bee changed). Occurs a taxa in a sediment core sample within the class 1, it might give better reconstructions, when the optimum for this specific species abundance class to a related environmental factor in the training set is used instead of the optimum of the hole response. This optimum can than be used to infer past environmental parameters. In general: Not the overall species optimum is used to infer environmental parameter, but several different optima were estimate related to different abundance classes using GLM with binomial distribution. The inferred value for the core sample is than calculated following the ML-Method: multiplying the class density functions of the species abundance class values from the core. The first version of this function (paltran 1.0) uses mainly the function polr from the package MASS (Venables, W.N. Ripley, B.D. (2002). Instead, each class is modelled using GLM. If a species occurs only rarely (<n) in the training set tow options are possible. First, it can removed (set.zero=TRUE), secondly the occurrence can add to the next lower class. On one hand this is a misclassification, on the other hand, the data are not completely lost, and might improve the model. If a species class from a core sample is not represent in the training set, the same options can be chosen. The species can bee removed (not.av="zero"), or the next lower density function can bee used. Thirdly the overall density function (product of all single class density functions) - assuming a symmetric species distribution - can be chosen (not.av="max"). As validation method until now only Leave One Out is available, as the algorithm is still very slow. With improving computing time, other methods like bootstrap might be incorporated.

Value

inferred train.set inferred environmental parameter for the training set
performance performance of the pom-regression
spec distribution species distribution curves for the single classes
inferred train.set (loo) inferred environmental parameter for the training set using leave one out as cross validation method
reconstruction_core.samples reconstructed environmental parameter for the samples of the core
mean (reconstruction_core.samples) (loo) reconstructed environmental parameter for the samples of the core using leave one out
sd (reconstruction_core.samples) (loo) standard deviation of the reconstructed environmental parameter for the samples of the core using "loo"

Author(s)

Sven Adler

References

in preperation

Examples

data(train_set.MV)
data(train_env.MV)
data(dud.df)
try<-mft(train_set.MV,train_env.MV,dud.df[1:3,],not.av="max",val=FALSE)

#using "loo" takes several minutes for computing


[Package paltran version 1.2-0 Index]